Writers of Pro Football Prospectus 2008

Most Recent FO Features

WentzCar16-1.jpg

» Week 11 DVOA Ratings

DVOA has finally climbed on board the Wentz Wagon! The Eagles move into the No. 1 spot, but they aren't the only strong, well-balanced team in the NFL this year. New Orleans, Pittsburgh, and the Los Angeles Rams make this one of the best seasons ever for multiple teams over 30% in DVOA, and Minnesota isn't far behind.

16 Oct 2015

NFL Injuries Part III: Variation by Position and Age

by Zach Binney

If you haven't read part I and part II of this series, I highly recommend it. Part I lays out what we're trying to do in this four-part series (describe NFL injuries) and gives some important caveats. Part II looks at how injury trends typically change throughout the season, and how they have changed over the years.

Today we have a whole post investigating the complexity of how injuries vary by age. Is a veteran really more likely to miss games as he ages? If so, how much?

Injury Risk by Age

The data in this section (except where indicated) excludes quarterbacks, kickers, punters, and other special teams players. I have done this because these players have lower baseline injury risks and make up a large proportion of players in older age groups (together they become a majority of players 37 and older). Including them would artificially deflate the injury risk for the oldest players.

The denominator for the chart and table below is player-seasons where the player played in at least one regular season game or was on IR the whole year. So all of these injury risks are only applicable for players who were still good enough to play if they weren't injured. A banged-up or washed-up 31-year-old linebacker doesn't make this chart unless there's a good chance he makes somebody's roster. He has to pass a smell test of "would he play if he were healthy?" before the injury risks below apply to him. Everything in the chart is conditional on passing that smell test.


Table 1. Number of Player-Seasons by Age
Age Player-Seasons 2000-14
20-21 172
22 1,256
23 3,111
24 3,790
25 3,479
26 3,017
27 2,544
28 2,091
29 1,760
30 1,413
31 1,078
32 786
33 548
34 328
35 206
36 115
37 61
38 31
39 13
40+ 13

Let's check out Figure 7. Your chance of getting listed on an injury report (blue line) rises steadily and substantially from about 56 to 59 percent at the age when most players are drafted (22 or 23 years old) to roughly 70 percent by 30, which is where it stays until around 37, when the vast majority of players have retired. Getting on the injury report doesn't mean you missed any time, but it can indicate you've had an injury that slowed you down, reduced performance, or decreased snaps.

What might be more relevant, though, is your risk of actually missing any games (orange line) or of missing significant time (at least four weeks, green line). Here we also see increases as players age, though they are smaller than most people would probably guess. For missing any games, we have a pretty steady risk of about 35 to 37 percent through age 29, with a spike to a new normal of about 40 to 42 percent after 30. It's a clear increase, but be honest: when you're afraid your team signed an aging veteran who is likely to get hurt, you were thinking the risk increase was more than 5 percent, weren't you?

For missing significant time (at least four weeks), the pattern is similar but the differences even smaller, jumping from 15 to 17 percent before age 30 to just under 20 percent after. That's only a 3 to 5 percent increase in risk.

What about those last few points on the chart, where injury risk plummets for those 38 and older? I've already excluded quarterbacks and special teams players, so that's not driving it. What is? It could be a few things:

  • Random error. From Table 1, we only have 31 player-seasons over 15 years for 38-year-olds, 13 for 39-year-olds, and 13 for those 40 and older. It's a small sample, so the drop could just be chance.
  • It could be that non-quarterback, non-special teams players who make it to 38 are just Iron Men with really low baseline injury risks. They might never miss a week before being forced to retire due to waning talent or athleticism. This is another example of the survivor bias that plagued last week's post, and we'll try to address it more below.
  • It could also be that when a player gets this old, trainers and coaches pay special attention to keeping him healthy: fewer snaps, less practice, etc. This could lower injury risk.

Effect of Categorizing the Oldest Ages Differently

I want to point out that the conclusions we might draw about injury risk in the oldest NFL players is remarkably sensitive to how we categorize age at the upper end of the distribution. Figure 8 presents the same data from Figure 7, except by defining the top age category as anywhere between 35-plus and 40-plus:

Depending on how we decide to categorize the oldest players, we could conclude that, for the oldest NFL players, there is A) basically no change in injury risk (top category 35-plus or 37-plus); B) a huge uptick in injury risk (top category 36-plus); or C) a precipitous drop in injury risk (top category 38-plus, 39-plus, or 40-plus).

All of these categorizations are defensible choices. I can reduce random error by upping my sample size if I lump everyone 35 and older together, which might be a fine idea if injury risk doesn't truly (i.e., biologically) vary between 35 and 40. However, that might be inappropriate if 39-year-olds aren't really like 35-year-olds with regards to their injury risk playing NFL football. Our data broken out into single years in Figure 7 would suggest that we shouldn't lump everyone 35 and older together, but the small sample sizes in those years leave me somewhat uncertain.

This is more a cautionary tale than anything else. It shows I can torture the data to make it confess whatever I want it to in certain cases. I guess what I'm saying is: don't trust anything I've written here. But seriously, when doing your own analyses, be cognizant of the effects seemingly mundane decisions can make on your conclusions, and test those effects in sensitivity analyses wherever possible.

Survivor Bias

I alluded to it in the second bullet above and described it in Part II, but a big caveat for this chart is survivor bias. You have to be pretty flippin' healthy to survive more than a few years in the NFL. When we get into older players, we're only looking at the cream of the crop, health-wise. By the time players hit their 30s, what's left is likely a group of Iron Men with a much lower baseline injury risk (and greater skill) than your average player. So if we just look at all 35-year-olds still playing in the NFL, it's not really fair to compare their injury risk with that of all 25 year-olds because the older players are, as a group, "heartier" than the younger players. This means the injury risk for older players is lower than it should be for a fair comparison.

What if we look at the same players over time, to get a better idea of how individual players age rather than the constantly-changing population of active players? We'll do that below.

But first, it's important to note that survivor bias isn't a big deal if Figure 7 is used properly. Remember, to be included you basically had to be likely to make a roster in a given season if healthy. As long as NFL coaches, general managers, and other personnel people make decisions on signing older players the same way they have for the last 15 years, those players' injury risks are probably fairly close to the historical averages in Figure 7. So while we may not be getting an accurate measure of the effect of aging on injury risk due to survivor bias, Figure 7 is still perfectly fine for assessing a player's injury risk once we have decided he's good enough to make a roster.

Of course, if Figure 7 revolutionized GM thinking and they started signing every injury-riddled 33-year-old they could find, I'm betting the injury risk of older players would rise. Our estimates incorporate, to a degree, conventional NFL wisdom on which older guys can still play, which likely keeps their observed injury risk down. If those judgments were to change, the risks could change dramatically.

Effects of Aging -- Watching the Same Players Over Time

OK, with that out of the way, let's eliminate survivor bias by looking at stable cohorts of players who played until at least a certain age in the NFL. All of the analyses below just consider the player's risk of missing one or more games in a year, since that's probably more important than just appearing on the injury report:

To get these stable cohorts we had to make a few exclusions. First we had to exclude anyone for whom we didn't have complete career data to ensure we had a stable cohort whose entire NFL careers we observed. We excluded all current players and anyone drafted before 2000 (when our data begins). We then excluded any player who retired (or otherwise left the NFL) prior to a given age. These exclusions took us from 6,373 non-QB, non-ST players from 2000-2014 to 2,275 (35.7 percent) with complete data who retired when they were 25 or older, 992 (15.6 percent) who retired at 28 or older, 522 at 30 or older, 200 at 32 or older, and 17 at 35 or older.

So what does Figure 9 tell us? It tells us that once we eliminate survivor bias we see a much stronger, more consistent, positive effect of age on injury risk. If we focus on the gray line (players who played in the NFL until they were at least 30) we see the risk of missing any time due to injury rising from around 30 percent in your early 20s to 50 or 60 percent by your mid-30s. That's a much larger rise than the roughly 5 percent increase we observed in Figure 7 over the same time period!

Part of the increased risk in older ages is probably driven by the increase in injuries in recent years (see Part II). We're looking here at long careers that began in 2000 or later, so older ages will correlate pretty strongly with recent years. This is especially likely since our risks at older ages in Figures 9 and 10 are substantially above those in Figure 7, which includes more older players from earlier years. I might try to tease out these effects in a separate post.

These risk increases grow more pronounced as we require players to have longer careers, which is exactly what we'd expect if survivor bias were at play in Figure 7. If you look at ages 23 to 25, you can see clearly that players who go on to retire later have lower injury risks when they're younger, suggesting they are a "heartier" group on average. Importantly, this could also partly be due to reverse causation -- if you got lucky with injuries when you were younger, you might go on to have a longer playing career. I'd suspect it's more that they're heartier, though.

One major flaw in Figure 9 is that players enter the NFL at different ages, so not every 25-year-old has the same NFL mileage on him. Let's tweak Figure 9 to replace age on the x-axis with years of NFL experience:

Phew, it's virtually identical. The left side of the chart is cleaner and still clearly shows lower injury risks for players with longer careers. We also see roughly the same magnitude of increases in injury risk as players accumulate more miles, though interestingly we don't get the violent spike at 11-plus years of experience as we did at 35-plus years of age in Figure 9. Admittedly this chart would be slightly improved by changing the categories from retirement ages to a minimum number of years played, but for now I wanted to maintain continuity with Figure 9.

For my money, I'd use Figure 10 to assess the effects of age on injury risk. However, if I'm a GM or fan just evaluating a potential acquisition's injury risk given his age, I'm still looking at Figure 7. Well, I'd actually look at more specific estimates by position and so on, but you get the idea.

Do Different Positions Age Differently?

I don't see an awful lot of variation in Figures 11A and 11B. The main thing I see is that linemen on both sides tend to have lower injury risks when they're younger, but then by their 30s they've caught up with their compatriots. That and, of course, that injury risks are substantially lower for special teamers (kickers and punters, mostly) than anyone else. I feel compelled to remind you they could all still beat me up, though.

Age Summary

If you want to estimate the injury risk for your team signing a player of a given age today in the NFL, use Figure 7. If you want an idea of how injury risk changes as an individual player ages, use Figure 10.

My overall take is that injury risk generally rises as players age, from somewhere around 30 percent when they enter the NFL to somewhere around 50 percent by their mid-30s (if they make it that long). I think we might currently over-discount some older players due to perceived injury risk, and there could be some value there for teams willing to take a chance or two on guys over 30.

This is another area where in-depth football knowledge from GMs, trainers, and others can really supplement analytics. It's hard to interpret this data properly without knowing a lot about the decision-making process for signing older players, which I'm the first to admit I do not.

Next Steps and Comments

In next week's installment we'll look more at how injuries vary by position. As always, comments welcome below!

Zach is a freelance injury analyst and a PhD student in Epidemiology focusing on predictive modeling. He consults for an NFL team and loves Minor League Baseball. He lives in Atlanta.

Posted by: Zach Binney on 16 Oct 2015

12 comments, Last at 13 Jan 2016, 7:12am by FlorenceReed

Comments

1
by Karl Cuba :: Fri, 10/16/2015 - 11:39am

Great article.

2
by dbostedo :: Fri, 10/16/2015 - 5:09pm

Agreed... loving these.

3
by PatsFan :: Fri, 10/16/2015 - 5:48pm

Definitely one of the better new things to appear on FO.

6
by Alternator :: Sat, 10/17/2015 - 6:04pm

Nothing really to say except that I agree here - it's interesting stuff to see, and it's the kind of thing that gives FO its character. Good work.

4
by Anon Ymous :: Fri, 10/16/2015 - 9:38pm

Great stuff.

5
by Tomlin_Is_Infallible :: Sat, 10/17/2015 - 12:09am

smh @ Excel, still.
--------------------------------------
The standard is the standard!

7
by armchair journe... :: Mon, 10/19/2015 - 9:56am

I am saving this article to help explain to students, so clearly, the difficulty of gleaning conclusions from data sets -- and ramifications of "seemingly mundane decisions." Fantastic.

//AJMQB

8
by Zach Binney :: Mon, 10/19/2015 - 2:27pm

Cool! I'm kinda curious who you envision using this for. As you might imagine I teach epi to grad students in various capacities.

9
by Tomlin_Is_Infallible :: Wed, 10/21/2015 - 9:02am

It could make for a very interesting multivariate regression or fitting/optimization exercise.

It's unfortunate that as presented the graphs are possibly not of sufficient quality
and clarity to motivate the students.

--------------------------------------
The standard is the standard!

10
by brian30tw :: Fri, 10/30/2015 - 9:38am

Agreed with everyone else: great series of articles.

Minor point, but I think there must be a calculation error with the grey line (Ages 36+) in Figure 8. That point should be a weighted average of all the points corresponding to ages 36 and above, where the weight is the number of player-seasons for each age. By construction, that weighted average must be less than the maximum and greater than the minimum value observed for those age groups, so it's impossible for the grey line to extend above the rest of the data. That doesn't change the overall point, though.

11
by Zach Binney :: Sun, 11/08/2015 - 10:23pm

Brian,

That's a great catch I'm embarrassed to say I didn't make. It was a spreadsheet coding error. In the interest of transparency (hooray, ethics), I want to clarify what Figure 8 should look like (and we'll try to get it changed soon):

Each line that's an older age group should have its final points dropping further and further. So 35+ and 36+ are flat, 37+ is a slight decline, and 38+ to 40+ are all precipitous drops. It doesn't change my central point much since the difference between flat risk and a precipitous drop is a pretty big deal, but my coding error did lead me to overstate the effects of this decision insofar as I can't actually make injury risk RISE by this categorization choice. For that, I apologize.

Thanks again for catching that!

12
by FlorenceReed :: Wed, 01/13/2016 - 7:12am

Position and Age changes according to time but to maintain it one have to take care of its health. And this can be done with proper guide related body health. This is very important for the sports person specially. cellublue.com is effective for this.